Long Short-Term Memory Recurrent Neural Network for Short-Term Freeway Traffic Speed Prediction
MetadataShow full item record
EXECUTIVE SUMMARY Traffic congestions on Riverside Freeway, California, SR 91 usually happen on weekdays from 0:00 to 23:55 of January 3, 2011 to March 16, 2012. Traffic congestions can be detected from average mainline speed, the free flow speed fluctuates around 60 to 70 mph, the congestion speed fluctuate around 30 to 40 mph. This research aims at predicting average mainline speed of the segment which is downstream to the merging road of a mainline and a ramp. The first objective is to determine the independent traffic features which can be used into the experiments, the second objective is to do data processing in order to form the data instances for the experiments, the third objective is to determine the predicted average speed with different time lags and to do the prediction, the fourth objective is to evaluate the performance of predicted average mainline speed. This research data source is the Caltrans Performance Measurement System (PeMS), which is collected in real-time from over 39000 individual detectors. These sensors span the freeway system across all major metropolitan areas of the State of California. PeMS is also an Archived Data User Service (ADUS) that provides over ten years of data for historical analysis. It integrates a wide variety of information from Caltrans and other local agency systems includes traffic detectors, census traffic counts, incidents, vehicle classification, lane closures, weight-in-motion, toll tags and roadway inventory. For this research, the experimental data is 280 weekdays from January 3, 2011 to March 16, 2012, traffic speed and flow of each lane of mainline, traffic flow of each lane of ramp, the average mainline speed, the total mainline flow and the total ramp flow which was collected by one mainline detector and one on-ramp detector every 5 mins. Previous studies proposed many different mathematical models, statistical models and machine learning models to predict traffic speed, flow and volume. The overall performance of the prediction is evaluated with overall accuracy values which measured with root-mean-square error (RMSE) and mean absolute percentage error (MAPE), besides, the predicted traffic speed is used to detect breakdown of congestions. However, the performance of existing traffic speed prediction models and methodologies has been criticized, many of the studies focus on simple data preparation and model application, traffic speed and flow are time series based data, the data set is very large, in order to improve the computation ability, feature selection, data processing and model selection are necessary for transportation research, in addition, the evaluation methods should also be considered to evaluate the performance of predictions. Upon analyzing the data for this research, two external effects were immediately evident, the first of which was analysis of the traffic data set to identify the independent, the data processing to generate traffic data instances, cross validation to generate training and test sets for the prediction experiments. The second external effect was the structure of the architecture of the Long short-term Memory Recurrent Neural Network (LSTM) for the future 5 to 30 min average mainline speed prediction. The presence of traffic congestions caused speed to fluctuate, rise or fall in ways, these traffic congestions were not identified to distribute in the same time range of a day through the analysis. It is difficult to capture the breakdown points of traffic congestions, especially the duration of traffic congestions is small, LSTM applications can capture most of the breakdown of traffic congestions even minor rapid fluctuation of average mainline speed across the 280 days. According to the research, most of the breakdown starting points of predicted average mainline speed can match the real breakdown starting points of average mainline speed, besides, the overall RMSE and MAPE results achieved 7.5 and 4.56 for future 30 min average mainline speed prediction.